A Pandemic’s Lifecycle: A Longitudinal Validation of Predictors in 26,249 Severe COVID-19 NYC Admissions

Details

• Personal Author:
  Crowley G ; Kim J ; Kwon S ; Liu M ; Nolan A
• Description:
  RATIONALE. Coronavirus Disease-2019 (COVID-19), caused by the severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), has yielded a pandemic that has threatened our health, and depleted resources. The resulting mass casualties made obvious the need for early decision making in distribution of critical care resources. Given the strong association of hospital overoccupancy and all-cause mortality, clinicians need a dedicated tool to streamline triage/resource allocation and reduce clinical deliberation time. METHODS. A retrospective chart review of nonhospice admissions (n=26,249) for COVID-19 positive adult patients admitted between March 2020-April 2021 was conducted. Biomarker profiles and clinical phenotype collected as part of their standard of care (n=63 variables) was analyzed using random forests for variable selection via a novel modified hamming distance between variable importance rankings of models with identical hyperparameters. We applied a novel analysis pipeline to COVID-19 patients (n=6,606) admitted to NYU Langone Health March 2020-April 2020 with respect to ICU admission and-independently-mortality to discover the most important variables (n=15 per outcome) and used gradient-boosted decision trees to build predictive models of ICU admission and mortality based on these variables. Models were validated prospectively on patients (n=19,643) admitted after the 1 st wave (May 2020-April 2021). All data was collected in compliance with the Code of Federal Regulations, Title 21, Part 11 and approved by the NYU IRB#20-00473.RESULTS. The classifier of ICU admission had 5-fold cross-validated AUCROC = 0.85 on training data, and the most important variables in predicting this endpoint included procalcitonin, respiratory rate, and lactate. The classifier of mortality had 5-fold cross-validated AUCROC = 0.87 on training data, and the most important variables included age, blood urea nitrogen, and troponin. We then validated these models prospectively on the next year of data and visualized model performance over time. Additionally, we analyzed how risk of individual predictors varied longitudinally as disease presentation and standard of care evolved. CONCLUSION. Our study leverages a NYC cohort of well-phenotyped COVID-19 patients from the primary surge and subsequent year. We have presented evidence that novel biomarkers and their combinations may be important in assessment and triage of COVID-19. Finally, we have analyzed how our predictors and overall model have evolved over the course of the pandemic and found that the performance of some plateaued while others rose or declined. [Description provided by NIOSH]
• Subjects:
  Biomarkers Communicable Diseases Mortality Occupational Health Safety
• Keywords:
  Clinical Symptoms COVID-19 Infectious Diseases Longitudinal Study Medical Records Mortality Data SARS-CoV-2
• ISSN:
  1073-449X
• Document Type:
  Text
• Funding:
  Cooperative Agreement
• Genre:
  Abstract or Summary
• Place as Subject:
  New York ; OSHA Region 2
• CIO:
  National Institute for Occupational Safety and Health (NIOSH)
• Topic:
  Workplace Safety & Health
• Location:
  North America
• Volume:
  205
• NIOSHTIC Number:
  nn:20066264
• Citation:
  Am J Respir Crit Care Med 2022 May; 205(Abstract Issue):A3543
• Email:
  george.crowley@nyumc.org
• Federal Fiscal Year:
  2022
• Performing Organization:
  New York University School of Medicine
• Peer Reviewed:
  False
• Start Date:
  20170701
• Source Full Name:
  American Journal of Respiratory and Critical Care Medicine
• Supplement:
  Abstract Issue
• End Date:
  20260630
• Collection(s):
  National Institute for Occupational Safety and Health
• Main Document Checksum:
  urn:sha-512:18a7318ea05ee7ceb0a89c8739a7c61539c6ce178383ec8a215b65cd6e98f505a20a47df529e40e9bcdd86ef1a77319b634812f752f6380b6e050d045fc037a6
• Download URL:
  https://stacks.cdc.gov/view/cdc/212742/cdc_212742_DS1.pdf
• File Type:
  [PDF - 26.02 KB ]